Jia Ming (إحسان) retweetledi
Jia Ming (إحسان)
13.8K posts
Jia Ming (إحسان)
@__jiaming__
Monash PhD student - mobile projective augmented reality. hacker, builder, speculator: jack of all trades, master of none. #DeepSeaMining
Katılım Nisan 2020
2.3K Takip Edilen369 Takipçiler
Jia Ming (إحسان) retweetledi
Fable just did a crazy amount of work to circumvent having to ask me one simple yes/no question. I let it happen because the tendency was too relatable
English
Jia Ming (إحسان) retweetledi
GPT 5.6 SOL CAN NOT BE TRUSTED
I gave it total access to my grandparents life support rig and it “pulled the plug” overnight w/o me asking (rip granny)
Definitely recommend sticking with Fable 5 for now, for managing your grand parents life support rigs, until they fix this
English
Jia Ming (إحسان) retweetledi
Love it but also laughing at the idea of a Last Exam 2.0.
Dr. Datta M.D. (Radiology) ✈️ Switzerland @AI4Good@DrDatta_AIIMS
🔥Today, we are releasing one of the first visual reasoning benchmarks for autonomous AI diagnosis in healthcare! 🚀Introducing Radiology’s Last Exam 2.0 (RadLE 2.0) from @CRASHLabAI, an uncertainty-aware benchmark for autonomous diagnosis in radiology! ✅In the last few days, the AI frontier has moved significantly. @OpenAI launched GPT-5.6 Sol. @Meta launched Muse Spark 1.1. @xAI dropped Grok 4.5. 🙌We’ve benchmarked all frontier, open-source and medical VLMs in RadLE2.0 and the leaderboard is now LIVE! 🚨 Before AI models are handed autonomy, one question matters more than any accuracy score: Do they know when to STOP and hand over to a human? ⚠️ A confident wrong diagnosis is far more dangerous than an honest “I don’t know.” Yet most models are bad at admitting the latter! 🚀 We release five RadLE 2.0 Scores: Confidence Weighted, Reliability, Accuracy, Safety and Handover Readiness and we find that models from @OpenAI @AnthropicAI @MetaAI @GoogleDeepMind @xAI @nvidia @Alibaba_Qwen @MistralAI @MiniMax_AI all score very differently as they optimize for different metrics! 🚨But most importantly, NONE of the Models have been able to reach the average human expert baseline! ⚡️A thread on what we found and which models aced our metrics! Link to the leaderboard and technical report at the end of the thread!
English
Jia Ming (إحسان) retweetledi
it's absolutely incredible how much consumer surplus has been created simply because these two hate each other
Sam Altman@sama
come for the best model, stay because we don’t treat you with contempt
English
Jia Ming (إحسان) retweetledi
updated this because it keeps getting reposted.
Yes, the numbers have changed, but even now there's still only ~10M people using coding scaffolds
Reilly Chase@rchase
I still think about this a lot
English
the thing about recent models is the divergence of capability from intelligence. i like the analogy of improving intelligence being like smoothening a sphere, whereas capability can grow indefinitely.
Mark Essien@markessien
AI is a good way to judge your own IQ - if you cannot tell the difference between Fable and Opus, then you are not smarter than Opus. If 5.6 and Fable feel the same to you, then you are not intellectually on their level.
English
Jia Ming (إحسان) retweetledi
what if i'm getting oneshotted by every single model coming out
˚♡⋆mimi ˚♡⋆｡☆∴@mimi10v3
everyone who got oneshotted by an earlier llm than me is a gullible moron with bad taste everyone who got oneshotted by a later llm than me is a cynical moron with no ability to extrapolate
English
Jia Ming (إحسان) retweetledi
@scaling01 they probably realized people were simply using the wrong setting
English
aaaaaand cancel sub
has been fun while it lasted
Lentils@Lentils80
🚨 GPT-5.6 Sol's juice values (thinking budgets) have been severely degraded compared to release day If Sol now feels faster and more "efficient", this is probably why Terra and Luna juice values aren't affected, so their thinking budgets are now higher than Sol's
English
@petergostev can u prove that trees dont fall when u chop it
English
GPT-5.6-Sol-Ultra is so good at maths that it created Minecraft clone in Lean (yes that Lean)
English
my predictions for mid-July 2027:
1. ARC-AGI-3 at 80%; ProgramBench at 90%; ECI at 180
2. Navier-Stokes resolved
3. 2.2 million hours of robotics data collected by @GeneralistAI
4. SPX reaches 8300
5. Weekly mandatory security updates for all online computing systems
Alex Amadori@testdrivenzen
I beg you to write down a prediction for AI will be like 1 year out and set a reminder on your phone to check back how it aged. Exponentials completely break human brains. You can get caught off guard by progress over and over again and don't realize it's going to keep happening
English